Attack behavior grade identification method based on improved reinforcement learning intelligent algorithm and folding shield thereof

Abstract

The invention relates to the technical field of anti-riot equipment, and discloses an attack behavior grade identification method based on an improved reinforcement learning intelligent algorithm and a folding shield thereof. Multi-modal sensor fusion, reinforcement learning and a real-time feedback control algorithm are applied to defense equipment, a protection strategy is optimized, the algorithm adapts to different working conditions and environment changes, prediction and early warning are performed on a future threat state level, and an intelligent algorithm is used for predicting and early warning the threat state level through combination of real-time data and historical data. And the prediction and planning capabilities of the system are improved. The system not only can quickly respond when a threat comes, but also can plan a protection strategy in advance to ensure that a protection action is ready before the threat occurs. The advanced planning reduces the response time, and ensures that the shield holder can cope with various sudden threats.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and more specifically, to an attack behavior level recognition method based on an improved reinforcement learning intelligent algorithm and a folding shield thereof. Background Art

[0002] In the current protection system, facing high-intensity and uncertain attacks, there are often the following main limitations:

[0003] Lack of real-time threat identification and classification: Traditional protection systems usually rely on fixed physical structures and manual operations when dealing with different threat levels. They are unable to prompt defenders to adjust protection strategies based on the characteristics or levels of the threats, resulting in delayed or inadequate responses in the event of sudden attacks.

[0004] Limited environmental perception capabilities: Existing protective equipment lacks multimodal perception methods to identify potential threats in complex environments. It usually relies on simple mechanical structures to resist attacks and lacks real-time monitoring and feedback of environmental changes such as surrounding temperature and sound, which affects the accuracy of its threat prediction and response.

[0005] Lack of adaptive learning and optimization: Traditional protection systems usually rely on preset fixed strategies and are unable to improve protection performance through continuous learning and optimization during use.

[0006] In addition, modern armed police, riot police and riot troops are often equipped with folding shields, so that they can resist attacks from hard objects, blunt objects and unknown liquids during riot control. However, in actual use, although the existing folding shields have the functions of pushing the opponent and protecting themselves, they cannot be extended due to their fixed size. Therefore, the protection range of the shield cannot be adjusted according to actual use needs, resulting in low flexibility in the use of the folding shield, reducing its applicability and bringing inconvenience to the operator. Therefore, it is necessary to improve it. Summary of the invention

[0007] In order to overcome the shortcomings of the prior art, the present invention provides an attack behavior level recognition method based on an improved reinforcement learning intelligent algorithm and a folding shield thereof.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] The attack behavior level recognition method based on the improved reinforcement learning intelligent algorithm includes the following steps: according to the environmental signals received from different sensors, multimodal sensor fusion, reinforcement learning and real-time feedback control algorithms are applied to the defense equipment to optimize the protection strategy, and the algorithm is adapted to different working conditions and environmental changes to predict and warn the future threat status level.

[0010] Further, the specific process is:

[0011] Step 1: Sensor data collection and preprocessing

[0012] Infrared sensor: used to detect the thermal radiation signal of the target and detect potential human or explosive danger sources; helps the system to issue an alarm in time when the target approaches, and judges the threat by calculating the temperature difference between the target and the environment;

[0013] Temperature change formula:

[0014] ΔT=T target -T ambient

[0015] Where ΔT is the temperature difference, T target : Target temperature, T ambient : Ambient temperature;

[0016] When the temperature difference ΔT exceeds the threshold set by the system, the sensor triggers an alarm, indicating the existence of a potential threat;

[0017] Sound sensor: used to detect sound wave signals in the environment, including gunshots, explosions and abnormal sounds; prompting users to adjust their defensive posture;

[0018] Fourier transform formula:

[0019] The Fourier transform converts the time signal x(t) into the frequency domain signal X(f), and the formula is as follows:

[0020]

[0021] Among them, x(t): sound waveform, time domain signal; X(f): frequency domain signal, which represents the energy distribution of the sound signal at different frequencies; f: this is the frequency, which represents the different frequency components in the sound signal; e -j2πft :This is the complex exponential function, which is the core of Fourier transform and is used to decompose the time domain signal into different frequency components; here j is the imaginary unit, indicating a complex number;

[0022] The key frequency components in the sound waves are extracted through Fourier transform and compared with the preset threat template. If the match is high, the system will sound an alarm.

[0023] Image capture module: used to collect real-time images through the camera, and use the optical flow method to help the system track the trajectory of high-speed moving objects in real time; the system will prompt the user to adjust the defensive posture according to the movement direction and speed of the threatening object;

[0024] The optical flow method is used to estimate the speed of objects in an image. The formula is as follows:

[0025]

[0026] Where, I: brightness value of the image; The rate of change of image brightness in the x and y directions; v x ,v y : The velocity components of the object in the x and y directions; The rate of change of image brightness over time;

[0027] By solving this equation, the system calculates the object's velocity vector v(x,y), thereby calculating the object's motion path and determining the object's threat level;

[0028] Acceleration sensor: used to detect changes in acceleration around the defense equipment, especially to identify and detect the impact of high-speed moving objects on the defense equipment, and prompt the defender to adjust the defense status of the defense equipment according to the changes in acceleration;

[0029] Newton's Second Law: The relationship between the acceleration a of an object and the force F and mass m applied to the object is:

[0030]

[0031] Among them, F: force; m: mass of the object; a: acceleration of the object;

[0032] By measuring the acceleration of an object, the object's trajectory and speed can be inferred; especially when a high-speed object approaches, the acceleration sensor can quickly capture its acceleration changes and issue an early warning;

[0033] Defense equipment provides appropriate installation locations for sensors. By integrating multiple sensors, the system captures multi-dimensional threat signals in the environment. To ensure the accuracy of the data, the sensor data must first be pre-processed. This includes:

[0034] Denoising: The sound of wind, background noise signals in sound data, and light noise in images are removed through filters; common methods include Gaussian filtering or median filtering to smooth data and enhance signal reliability;

[0035] Smoothing filtering: Acceleration and temperature sensor data often experience instantaneous fluctuations; the system smoothes these data through weighted moving average or Gaussian filter to reduce the impact of fluctuations on threat judgment;

[0036] The weighted moving average formula is:

[0037]

[0038] in, is the smoothed data at time t, w iis the weighting coefficient at each time point, x t-1 is the original data at time ti;

[0039] Feature extraction: Fourier transform is used to extract frequency features from sound signals, while optical flow is used to extract motion features from images. The multimodal data after feature extraction will be input into the subsequent self-attention mechanism for further analysis.

[0040] Step 2: Data processing using self-attention mechanism

[0041] The self-attention mechanism is responsible for processing data from infrared, sound, image, and acceleration multimodal sensors. By assigning different weights to different signal sources, the system prioritizes the most important threat signals. This dynamic adjustment mechanism ensures that the system responds to threats in real time and takes protective measures in complex environments.

[0042] The core of the self-attention mechanism is to calculate the attention weight of each input signal, which is used to judge the importance of the signal and ensure that the system can prioritize the signal with the greatest threat level, which is used to determine the system's protection response.

[0043] The system receives data x1, x2, ..., x from different sensors n , where each signal x i Represents the input of a sensor; the self-attention mechanism will assign a weight α to each signal i , indicating the priority of the signal; the weight calculation formula is:

[0044]

[0045] Where n, j is the number of signals, exp(e i ) means e i The expectation, α i : signal x i The attention weight ranges from 0 to 1, indicating the priority response degree of the signal; i : signal x i The relevance score of the signal x indicates the threat level of the signal to the current situation; the score is determined by the signal x i W is the weight matrix learned by the system during the training process, which is used to measure the impact of each signal source on threat assessment.

[0046] The Softmax function is used to ensure that the sum of the weights of all signals is 1, which means that the system will normalize all sensor inputs so that different signals can be responded to according to priority; the attention weight is determined by the relevance score e i It is decided that signals with greater threat levels will be given higher weights and thus processed first;

[0047] Another key feature of the self-attention mechanism is dynamic weight adjustment; as the environment changes in real time, the system adjusts its weights according to changes in different sensor input signals. In particular, when specific threat signals are detected, the system automatically increases the priority of these signals.

[0048] Signal Input: The system receives input signals from multimodal sensors:

[0049] x1: The infrared sensor detects a sharp rise in temperature;

[0050] x2: The sound sensor detects abnormal sound waves;

[0051] x3: The image sensor recognizes an object approaching at high speed;

[0052] x4: The acceleration sensor detects the acceleration changes around the defense device;

[0053] Weight calculation: The system calculates the relevance score e based on the characteristics of each signal i and the attention weight α i ;

[0054] Dynamic update of weights: According to the formula The system dynamically adjusts the weight of each signal;

[0055] Dynamic adjustment: When the image sensor detects an object approaching at high speed, the system will automatically increase the priority of the signal, and the defense equipment will be deployed first for protection; the weight setting is updated at a high frequency to ensure that the system can respond to changes in the external environment in real time;

[0056] When the image sensor detects a high-speed object approaching, the system will respond to this threat first, command the defense equipment to quickly deploy the protection direction to align with the threat source, and expand the baffle to the maximum protection area in a very short time;

[0057] Infrared sensors continuously monitor changes in ambient temperature. When the temperature rises, the system will adjust its protection strategy to deal with explosion or fire threats.

[0058] When the sound sensor detects abnormal sound waves, the system will further determine whether the threat comes from gunfire or other abnormal sounds, and make auxiliary judgments;

[0059] Through the self-attention mechanism, the system dynamically adjusts the priority of sensor data and combines the physical structure of the defense equipment to provide effective protection.

[0060] Step 3: Time series analysis for sequence data analysis

[0061] The time series analysis model is the core tool for analyzing time series data. Through the self-attention mechanism, the time series analysis model can identify the patterns in the historical data of the sensor and predict the future threat development trend. This mechanism is particularly suitable for multimodal data processing in dynamic environments. Defense equipment needs to predict threats in advance by analyzing acceleration, image, and sound sensor data. The following is a detailed description of this step.

[0062] The time series analysis model is constructed by inputting the time series X = [x1, x2, ..., x t ], to predict the output sequence Y = [Y1, Y2, ..., Y t ]; the input sequence X represents the historical data of acceleration, image, sound, and temperature obtained by the system from multiple sensors, and the output sequence Y is the system's prediction of the object's future position, speed, impact time, and future threats;

[0063] Signal input: The time series analysis model can capture long-term dependencies in the time series, enabling the system to predict future threat signals based on historical data; its processing process includes the following steps:

[0064] Signal input: The time series X obtained by the system = [x1, x2, ..., x t ], each x i represents the sensor input at a specific time step;

[0065] x1: data captured by the acceleration sensor 1 second ago;

[0066] x2: object position data captured by the image sensor 2 seconds ago;

[0067] x3: comprehensive input of sensors such as image and sound at the current moment;

[0068] Calculation of query, key, and value matrices: Time series analysis generates query Q, key K, and value V matrices by encoding the input signal at each time step:

[0069] Query matrix Q: used to predict future state questions; current object acceleration, velocity and other data are encoded into the query matrix to inquire about future threat developments;

[0070] Key matrix K: saves historical signal features to help identify past data related to the current signal; the historical object motion trajectory is encoded into the key matrix for current matching;

[0071] Value matrix V: stores the historical data values ​​corresponding to the key matrix for the final prediction calculation;

[0072] Weighted calculation and output: Time series analysis is done by multiplying the query matrix Q with the transpose of the key matrix K and then scaling it by a factor d k Adjustments are made to calculate the correlation between the current signal and the historical signal; the higher the correlation, the more important the historical signal is to the current prediction; the formula is as follows:

[0073]

[0074] Where: Q: represents the question raised by the current input signal when predicting future threats; this matrix is ​​generated by encoding the features of the sensor data in the current time step; K: represents the signal data in the historical time step, which is used to match with the query matrix to help the system determine which historical data is relevant to the current signal; QK T : Calculate the matching degree between the current signal and the historical data; V: represents the eigenvalues ​​of the historical data that matches the key matrix, and the system uses these data to infer the future state; d k : Scaling factor, used to avoid numerical instability caused by excessive input signals; usually the dimension size of the query and key matrices; softmax function: normalizes the calculation results to ensure that the weight distribution is reasonable and the sum of the correlations is 1;

[0075] This formula calculates the correlation between the query matrix Q and the key matrix K to determine which historical signals best reflect the threat changes in the current environment, thereby using historical data to predict future threat states; by scaling factor d k , the calculation results are stabilized;

[0076] In the protection system of defense equipment, the time series analysis model analyzes the time series data from multiple sensors to help the system predict future threats in advance; the acceleration sensor and image sensor jointly provide the movement information of the object. The system uses this data to determine the impact time and path of the object and promptly prompt to deploy protection;

[0077] The accelerometer measures the acceleration of an object and provides historical data about the object's motion state. By analyzing this data, the time series analysis model predicts the object's future position. If the acceleration of an object continues to increase, the system will predict that the object will hit the defense equipment at some point in the future, thereby deploying the defense equipment in advance.

[0078] The image sensor captures the position changes of the object and combines it with the time series analysis to predict the object's trajectory and final landing point. When the system detects that an object is a high-speed projectile, the time series analysis infers the object's future path through historical data and automatically prompts the defender to adjust the protection direction of the defense equipment to ensure that the protection action is completed before the object contacts the defense equipment.

[0079] Step 4: Reinforcement learning to optimize defense strategy

[0080] The core goal of reinforcement learning is to optimize the defense behavior of the protection system by maximizing the cumulative expected reward; the defense strategy of the protection system is π, and its expected reward is defined by the following formula:

[0081]

[0082] Among them, J(π): represents the expected return of strategy π, that is, the total cumulative reward obtained after the system executes the strategy; r t : The immediate reward at time step t, which depends on the current defensive behavior and the threat level of the environment; γ: A discount factor ranging from 0 to 1, used to balance the weights of current rewards and future rewards; a larger γ emphasizes long-term rewards, while a smaller γ focuses more on short-term gains;

[0083] In order to optimize the defense strategy, reinforcement learning uses the policy gradient method to update the policy parameter θ, and its formula is:

[0084]

[0085] in, Represents strategy π θ The gradient of π is used to guide the adjustment of the strategy parameter θ to improve the defense effect of the system; θ (a t |s t ): The system is in state s t Take action a t The probability of R is determined by the current strategy parameter θ; t : The cumulative reward starting from time step t, reflecting the sum of the defense effects of the system from the current time step to the future time;

[0086] Through this formula, the protection system continuously adjusts the strategy parameter θ to optimize its defense response, ensuring that the optimal protection strategy is always adopted in different threat scenarios;

[0087] Specific implementation steps of reinforcement learning to optimize defense strategy

[0088] Definition of state and action space: In order for the protection system to effectively respond to external threats, it is first necessary to define the state space and action space of the system;

[0089] State space S: Obtain environmental information through the system's multimodal sensors, including:

[0090] Acceleration sensor: provides acceleration data of the target object to help determine the movement trend of the object;

[0091] Image sensor: captures the motion trajectory of the threatening object and determines its position and direction of movement;

[0092] Sound sensor: detects abnormal sounds in the environment, such as explosions and gunshots, and determines the occurrence of threats;

[0093] Temperature sensor: monitors ambient temperature and identifies potential fire or explosion threats;

[0094] Action space A: defines the protection actions taken by the protection system, including:

[0095] Maximum protection state: the defense equipment is fully deployed to provide the largest area of ​​protection, suitable for high-threat scenarios;

[0096] Compact state: The defense equipment is partially retracted to maintain basic protection while saving energy, which is suitable for low-threat scenarios;

[0097] Quickly adjust the angle: adjust the angle of the defense device according to the threat signal detected by the sensor so that it can face the threat object head-on;

[0098] Reward function design: Reinforcement learning through reward function r t Guide the system to learn the best defense strategy; the reward function should be designed according to different threat scenarios to ensure that the system can make reasonable defense decisions in complex environments;

[0099] High Threat Protection Success Reward: When the system successfully defends against high-speed projectiles or explosives, the system will receive a higher immediate reward. t ,encourages the system to adopt the same protection strategy when encountering similar situations in the future;

[0100] Low-threat energy-saving rewards: In low-threat situations, if the system remains compact and reduces unnecessary defense deployment, positive rewards will be obtained to encourage energy-saving behavior;

[0101] Defense failure penalty: If the system fails to deploy defense in time, resulting in the threat hitting the defender, the system will receive a negative reward to reduce similar defense failures;

[0102] Action selection and strategy optimization: At each time step t, the system selects the action based on the current state s. t Choose the best defensive action a t ; The system adopts strategy π θ (a t |s t ) controls the action selection and uses the policy gradient method to update the parameter θ to ensure the maximum cumulative reward in future threat situations;

[0103] Continuous learning and dynamic adjustment: A key advantage of reinforcement learning is that it can accumulate experience through continuous environmental interaction; over time, the system gradually accumulates more defense experience and continuously optimizes its strategy, enabling it to handle a wider variety of threat scenarios;

[0104] High Threat Response: When the system detects through sensors that an object is approaching quickly and is highly threatening, the system will quickly deploy defense equipment and adjust the angle to maximize the defense against impact;

[0105] Low-threat response: In a low-threat environment, the threat detected by the sensor is small, and the system will choose to maintain a compact state, reduce energy consumption and extend the system's operating time;

[0106] Reinforcement learning combines sensor input and environmental threats to dynamically adjust protection actions in the protection system. Through continuous learning, the system optimizes the deployment status and protection angle of defense equipment based on sensor input information to achieve efficient and intelligent protection.

[0107] High-threat scenario: When the sensor detects a threatening object approaching at high speed, the system analyzes the threat level through a reinforcement learning algorithm and selects the maximum protection strategy based on historical protection records. At this time, the defense equipment will be deployed quickly to ensure the widest protection area and reduce the risk of attack to the defender.

[0108] Low-threat scenarios: In low-threat situations, when environmental changes are small and objects move slowly over long distances, the system will maintain a compact state through a low-energy strategy to save energy consumption; this decision guides the system through a reward mechanism, allowing the system to reduce unnecessary protection operations in low-threat situations, thereby extending the working time of the defense equipment.

[0109] Furthermore, the defense device is a shield, and the protection system includes a controller arranged on one side of the shield to execute an algorithm, and sensors and actuators connected to the controller, the sensors include sound sensors, temperature sensors, acceleration sensors, and image sensors; the actuators include vibration motors and voice prompt speakers.

[0110] Further, the method further comprises the following steps:

[0111] Step 5: Real-time feedback and defensive action adjustment

[0112] The real-time feedback mechanism is regarded as a feedback control system, which is used to make adjustments based on the threat signal input by the sensor and the current defense state of the system. Let the current defense state be S t , the feedback adjustment formula is:

[0113] S t+1 =S t +α(r t -St )

[0114] Among them, S t : represents the defense state of the system at time t; the system is in different defense states such as fully deployed, defense angle adjustment, and compact state; α: learning rate, ranging from 0 to 1; a larger α value means that the system responds to new threat signals more quickly, while a smaller α value means that the system will gradually adjust and be more adaptable to prevent frequent switching; r t : The feedback signal at the current time t, reflecting the threat level detected by the system from the sensor input information; this feedback signal is based on information such as object acceleration, velocity, and temperature change; S t+1 : represents the defense state of the system at the next time step t+1;

[0115] When the system detects new threat information, t Will be updated according to the current defense status S t and threat feedback t The formula calculates the difference between the next moment S t+1 Protection adjustment strategy; This feedback formula draws on the feedback control principle in control theory, so that the system can gradually move towards the optimal protection state;

[0116] Steps to implement the real-time feedback mechanism

[0117] Sensor data input: The system relies on multi-modal sensors to collect data in real time and assess the threat level based on this data, as follows:

[0118] Accelerometer: detects objects moving at high speeds, indicating that potential threats are approaching the shield quickly;

[0119] Image sensor: Identifies the direction, speed, and size of an object, helping the system determine the threat level of the object;

[0120] Sound sensor: detects abnormal sounds and indicates that the shield protection system is facing threats from sound waves;

[0121] Temperature sensor: monitors drastic changes in ambient temperature, high temperature explosion or fire source, and prompts the system to face extreme danger;

[0122] Real-time adjustment of defense status: The system adjusts the defense status based on the threat level signal fed back by the sensor and the current defense status S t , use the formula to dynamically adjust the shield status; the specific adjustments include:

[0123] Fully deploy the shield: When the threat level is high, the system will prompt the operator to fully deploy the shield to provide maximum protection and prevent high-speed approaching objects from hitting the shield holder;

[0124] Compact State: In low-threat situations, the system keeps the shield in a compact state, saving energy and avoiding unnecessary deployment operations;

[0125] Protection angle adjustment: When the sensor identifies a threatening object coming from a specific direction, the system will prompt the operator to adjust the shield angle to face the threatening direction at the best angle to reduce the exposed area;

[0126] Real-time response and learning: Through real-time feedback mechanism, the system can continuously adjust the shield status to cope with the dynamically changing threat environment; reinforcement learning strategy supports the system to gradually optimize the protection strategy in different scenarios;

[0127] Quickly adjust the protection angle: When the sensor detects an object approaching at high speed from a certain angle, the system will quickly adjust the protection angle and deploy the shield accordingly;

[0128] Gradually optimized protection strategy: In the ever-changing threat scenarios, the system gradually optimizes the protection action through sensor feedback signals to ensure that the shield holder is always in the best protection state;

[0129] Step 6: Shield Bearer's Prompt System

[0130] The shield holder's reminder system is a core module of the shield protection system. It reminds the shield holder of the current defense status and danger level through vibration and sound prompt feedback. This design ensures that the shield holder can quickly and accurately take corresponding defense measures when facing complex and dynamic threats. The following is a detailed implementation plan, and gives specific implementation cases and descriptions of related parameter variables;

[0131] The prompt system provides feedback to the shield holder through the intensity of vibration prompts and sound prompts based on the detected danger level L; as the danger level changes, the feedback intensity will also be adjusted to ensure that the shield holder can take appropriate actions according to different danger situations; the vibration prompt intensity formula is:

[0132] V=kL

[0133] Where V: the intensity of the vibration prompt, which indicates the vibration feedback received by the shield holder; as the danger level increases, the intensity of the vibration prompt also increases; L: the danger level detected by the current system, ranging from 0 to 1, 0 represents no threat, and 1 represents the highest threat; k: the proportional coefficient, used to adjust the sensitivity of the vibration prompt; a higher k value means that the vibration intensity will increase significantly when the danger level changes slightly; the sound prompt frequency formula is:

[0134] f=f0+k ′ L

[0135] f: The frequency of the sound prompt, which is used to remind the shield holder of the current threat level through the sound alarm; as the danger level increases, the frequency of the sound prompt will also increase; f0: The basic frequency, which indicates the frequency of the sound prompt when there is no threat; the shield holder perceives the change of danger level through the change of sound frequency; k ′ : Adjustment coefficient, which determines the influence of danger level on the frequency of sound prompts; higher k ′ The value means that when the danger level increases, the frequency change of the sound prompt becomes more obvious;

[0136] Implementation steps of the shield bearer reminder system

[0137] Sensor data input: The prompt system obtains the current danger level L from multiple sensors and feeds it back to the shield holder in real time; the following are the relevant sensor inputs:

[0138] Acceleration sensor: detects the speed and acceleration of an object to determine whether a threatening object is approaching quickly;

[0139] Image sensor: captures the movement trajectory, direction and distance of threatening objects to help the system determine the threat level of the object;

[0140] Sound sensor: detects abnormal sounds in the environment and indicates potential threats;

[0141] Temperature sensors: monitor ambient temperature changes, especially when detecting fire or explosives;

[0142] Vibration prompt: When the system detects a threat, it provides feedback to the shield holder through vibration prompt; the vibration intensity is adjusted according to the danger level L:

[0143] When the threat level is high: the system will send a strong vibration prompt to the shield holder, reminding the shield holder to take defensive measures immediately;

[0144] At low threat level: the system will only provide slight vibration to remind the shield holder that the current threat is low; at this time, the V value is small and the vibration feedback is limited;

[0145] Sound prompt: At the same time, the system issues an alarm through the sound prompt module, and the frequency f of the sound prompt increases as the danger level L increases:

[0146] When the threat is high: the frequency of the sound prompt increases significantly, reminding the shield holder that he is facing a serious threat; as the danger level L increases, the sound becomes more urgent;

[0147] When the threat is low: the sound prompt frequency is kept at a low level f0, reminding the shield holder that he is in a relatively safe state;

[0148] Prompt feedback and defensive action adjustment: The prompt system is combined with the shield's defensive action to ensure that the shield holder can quickly make corresponding adjustments based on the feedback provided by the prompt system; when the shield holder receives high-threat vibration and sound prompts, take defensive measures immediately, such as:

[0149] Adjust the shield's protection angle to deal with threats from different directions;

[0150] Fully deploy the shield to deal with high-speed or high-risk threats;

[0151] Actively reposition to avoid head-on impact.

[0152] The present invention discloses a foldable shield that is easy to extend, comprising:

[0153] A shield body, wherein explosion-proof glass is fixedly installed inside the shield body, a square groove is opened at the bottom end of the front side of the shield body, and a long groove is opened at the bottom end of the shield body;

[0154] An extension mechanism, wherein the extension mechanism is arranged inside the long slot;

[0155] Wherein, the extension mechanism includes a telescopic baffle, the outer surface of the telescopic baffle is movably connected to the inside of the long slot, a square block is fixedly installed on the top of the telescopic baffle, the outer surface of the square block is movably connected to the inside of the long slot, a first spring is fixedly installed on the top of the square block, the top of the first spring is fixedly connected to the inside of the long slot, a tooth plate is fixedly installed on the front of the telescopic baffle, a block is movably connected to the front of the tooth plate, the other end of the block extends to the front of the front of the shield body through the square slot, the outer surface of the block is movably connected to the inside of the square slot, a fixed block is movably sleeved inside the block, the back of the fixed block is fixedly connected to the front of the shield body, a second spring is fixedly installed on the top of the back of the block, and one end of the rear of the second spring is fixedly connected to the front of the shield body.

[0156] As a preferred technical solution of the present invention, round rods are fixedly installed on the left and right sides of the shield body, the top and bottom ends of the outer surfaces of the round rods are movably sleeved with round tubes, the outer surface of the round tubes is fixedly installed with folding baffles, and the back side of the folding baffles is movably connected to the front side of the shield body.

[0157] As a preferred technical solution of the present invention, a fixing plate is fixedly installed on the top of the front side of the shield body, vertical rods are fixedly installed on the left and right sides of the top of the fixing plate, and long rods are movably sleeved on the outer surfaces of the vertical rods.

[0158] As a preferred technical solution of the present invention, a third spring is fixedly installed at the bottom end of the long rod, and the bottom end of the third spring is fixedly connected to the top end of the fixed plate.

[0159] As a preferred technical solution of the present invention, a first round block is fixedly installed on both sides of the long rod, the bottom end of the first round block is movably connected to a second round block, the bottom end of the second round block is fixedly connected to the top end of the round tube, and the bottom end of the second round block is movably connected to the top end of the round rod.

[0160] As a preferred technical solution of the present invention, a square plate is fixedly mounted on the front of the shield body, and vertical grooves are provided at both upper and lower ends of the left and right sides of the square plate.

[0161] As a preferred technical solution of the present invention, a threaded rod is movably sleeved inside the square plate, and the top and bottom ends of the threaded rod both penetrate the interior of the square plate and extend to the top and bottom ends of the square plate respectively.

[0162] As a preferred technical solution of the present invention, a moving block is threadedly sleeved on the outer surface of the threaded rod, and the front surface of the moving block is movably connected to the inside of the square plate.

[0163] As a preferred technical solution of the present invention, movable blocks are fixedly installed on the left and right sides of the moving block, and the internal movability sleeve of the movable block is connected with a moving rod. There are two moving rods, and the other ends of the two moving rods extend to the front of the square plate through vertical grooves, and the outer surfaces of the other ends of the two moving rods are movably sleeved with an extrusion plate.

[0164] As a preferred technical solution of the present invention, a rotating block is fixedly installed on the top end of the threaded rod, and the bottom end of the rotating block is movably connected to the top end of the square plate.

[0165] Compared with the prior art, the present invention has the following beneficial effects:

[0166] 1. The present invention provides a telescopic baffle, a first spring, a toothed plate and a stopper. When the stopper rotates with the movable sleeve joint with the fixed block as the axis, the stopper will release the obstruction to the overall movement of the toothed plate during the rotation. Due to the elastic force of the first spring, the first spring will drive the telescopic baffle to move as a whole through the square block. Due to the limiting effect of the long groove, the telescopic baffle will move downward. Subsequently, the overall protection area of ​​the shield body will increase during the downward movement of the telescopic baffle, thereby achieving the effect of facilitating the extension of the shield body.

[0167] 2. The present invention sets a threaded rod, a motion block, an active block and a motion rod. Since the interiors of the two motion blocks are threadedly sleeved with the outer surfaces of the threaded rods, when the threaded rods rotate along the interior of the square plate, the two motion blocks will move. Since the outer surfaces of the two motion blocks are movably sleeved with the interior of the square plate, the two motion blocks will drive the four active blocks to move in opposite directions. Subsequently, the two motion rods will move driven by the four active blocks. At this time, the other ends of the two motion rods will drive the extrusion plate to move backward. At this time, the extrusion plate will cooperate with the square plate during the backward movement to clamp and fix the operator's hands, thereby preventing the operator from losing their hands when using the shield.

[0168] 3. Enhanced adaptive threat identification and response capabilities: After the introduction of multimodal sensor fusion (including infrared, sound, acceleration and image sensors), the system has stronger perception capabilities. Different sensors capture environmental information in different dimensions. Combined with the self-attention mechanism and data fusion algorithm, the system can more accurately determine the source, nature and level of threats. For example, infrared sensors (temperature sensors) are used to capture heat sources to determine whether there is a person approaching or an explosion threat; sound sensors identify abnormal sound waves, such as gunshots and explosions, analyze their spectral characteristics, and determine the source of danger. Acceleration and image sensors can be used to track the trajectory of flying objects in real time. The data of multimodal sensors is processed by algorithms, which can accurately identify threats in complex environments, reduce false alarms, and enhance overall response capabilities.

[0169] 4. Real-time dynamic adjustment of protection strategy: Through reinforcement learning and real-time feedback control mechanism, the system can automatically adjust the protection strategy according to the real-time changes in the threat level. In actual operation, the reinforcement learning algorithm continuously optimizes the system decision through the reward mechanism to ensure that the system makes a rapid protection response under high threats and maintains energy conservation under low threats. The system's protection strategy is no longer fixed, but dynamically adjusted. For example, when the danger level rises, the system will prompt the defender to expand the shield to the maximum protection state, and shrink it to a compact state when the danger level decreases.

[0170] 5. Improve the robustness and stability of the protection system: Over time, the system accumulates data and optimizes its protection strategy through continuous reinforcement learning and adaptive algorithms. This enables the system to remain robust and stable when facing diverse and complex threat environments. Even in harsh combat environments, the system can reduce false alarms through self-optimization algorithms and improve its stability and effectiveness.

[0171] 6. Enhanced threat prediction capabilities: Through time series analysis algorithms, such as the Transformer model, the system can extract threat patterns from historical sensor data and predict the development trend of future threats. This mechanism helps the system make predictions before threats occur, prompting defenders to take protective actions in advance. For example, the system can predict the future impact time and direction of an object based on its acceleration and trajectory data, and be ready for protection before the threat actually arrives.

[0172] 7. Improve the shield holder's operating experience: Through the integrated prompt module (including vibration and sound feedback), the shield holder can grasp the current defense status and threat level in real time, avoiding delayed response due to inaccurate judgment. When the danger level increases, the system will remind the shield holder through strong vibration and high-frequency sound alarm to take protective measures in time. This intelligent prompt greatly improves the shield holder's combat efficiency, especially in complex environments, without distracting attention from monitoring equipment, but understanding the system status through direct physical feedback. BRIEF DESCRIPTION OF THE DRAWINGS

[0173] Figure 1 It is a schematic diagram of the structure of the present invention;

[0174] Figure 2 It is a schematic diagram of the back structure of the present invention;

[0175] Figure 3 It is a cross-sectional structural schematic diagram of the present invention;

[0176] Figure 4 It is a schematic cross-sectional structure diagram of a round rod of the present invention;

[0177] Figure 5 It is a structural schematic diagram of the tooth plate of the present invention;

[0178] Figure 6 It is a bottom view structural schematic diagram of the present invention;

[0179] Figure 7 for Figure 1 A schematic diagram of the local enlarged structure at point A in the middle;

[0180] Figure 8 for Figure 3 A schematic diagram of the local enlarged structure at B in the middle;

[0181] Fig. 9 : Flowchart of the intelligent algorithm introduced in the present invention.

[0182] In the figure: 1. shield body; 2. explosion-proof glass; 3. square groove; 4. long groove; 5. telescopic baffle; 6. square block; 7. first spring; 8. tooth plate; 9. baffle; 10. fixed block; 11. second spring; 12. round rod; 13. round tube; 14. folding baffle; 15. fixed plate; 16. vertical rod; 17. long rod; 18. third spring; 19. first round block; 20. second round block; 21. square plate; 22. vertical groove; 23. threaded rod; 24. moving block; 25. movable block; 26. moving rod; 27. extrusion plate; 28. rotating block. DETAILED DESCRIPTION

[0183] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0184] like Figures 1 to 8 As shown, the present invention provides a foldable shield that is easy to stretch, comprising:

[0185] A shield body 1, an explosion-proof glass 2 is fixedly installed inside the shield body 1, a square groove 3 is opened at the bottom end of the front of the shield body 1, and a long groove 4 is opened at the bottom end of the shield body 1;

[0186] A stretching mechanism, the stretching mechanism is arranged inside the long slot 4;

[0187] Among them, the extension mechanism includes a telescopic baffle 5, the outer surface of the telescopic baffle 5 is movably connected to the inside of the long slot 4, a square block 6 is fixedly installed on the top of the telescopic baffle 5, the outer surface of the square block 6 is movably connected to the inside of the long slot 4, a first spring 7 is fixedly installed on the top of the square block 6, the top of the first spring 7 is fixedly connected to the inside of the long slot 4, a tooth plate 8 is fixedly installed on the front of the telescopic baffle 5, a block 9 is movably connected to the front of the tooth plate 8, the other end of the block 9 extends to the front of the front of the shield body 1 through the square slot 3, the outer surface of the block 9 is movably connected to the inside of the square slot 3, the internal movably sleeved with a fixed block 10, the back of the fixed block 10 is fixedly connected to the front of the shield body 1, and a second spring 11 is fixedly installed on the top of the back of the block 9, and one end of the rear of the second spring 11 is fixedly connected to the front of the shield body 1.

[0188] When the stopper 9 rotates at the movable socket joint with the fixed block 10, the stopper 9 will release the blocking effect on the overall movement of the tooth plate 8 during the rotation. Due to the elastic force of the first spring 7, the first spring 7 will drive the telescopic baffle 5 to move downward along the inside of the long slot 4 through the square block 6, thereby extending the shield body 1 as a whole. Due to the elastic force of the second spring 11, the movement of the stopper 9 will have a good resetting effect.

[0189] Among them, round rods 12 are fixedly installed on the left and right sides of the shield body 1, and round tubes 13 are movably connected to the top and bottom ends of the outer surface of the round rods 12. A folding baffle 14 is fixedly installed on the outer surface of the round tube 13, and the back side of the folding baffle 14 is movably connected to the front side of the shield body 1.

[0190] Since the insides of the four round tubes 13 are movably connected to the outer surfaces of the two round rods 12 respectively, the two folding baffles 14 can rotate around the movably connected parts of the four round tubes 13 and the two round rods 12 as the axis, so as to unfold the two folding baffles 14.

[0191] A fixing plate 15 is fixedly mounted on the top of the front face of the shield body 1 , vertical rods 16 are fixedly mounted on both left and right sides of the top of the fixing plate 15 , and a long rod 17 is movably sleeved on the outer surface of the vertical rod 16 .

[0192] Due to the design of the two vertical rods 16 , the movement of the long rod 17 is limited, so that the long rod 17 can only move up and down along the outer surfaces of the two vertical rods 16 .

[0193] The third spring 18 is fixedly mounted on the bottom end of the long rod 17 , and the bottom end of the third spring 18 is fixedly connected to the top end of the fixing plate 15 .

[0194] When the long rod 17 moves upward, the third spring 18 will be in a stretched state. Due to the elastic force of the third spring 18, the movement of the long rod 17 will be well reset.

[0195] Among them, the first round block 19 is fixedly installed on the left and right sides of the long rod 17, the bottom end of the first round block 19 is movably connected to the second round block 20, the bottom end of the second round block 20 is fixedly connected to the top of the round tube 13, and the bottom end of the second round block 20 is movably connected to the top of the round rod 12.

[0196] When the first round block 19 contacts the second round block 20, the first round block 19 will lock the overall rotation of the round tube 13 and the folding baffle 14 through the second round block 20. When the long rod 17 drives the two first round blocks 19 to move upward, the two first round blocks 19 will release the locking effect on the two second round blocks 20.

[0197] A square plate 21 is fixedly mounted on the front of the shield body 1 , and vertical grooves 22 are provided at both upper and lower ends of the left and right sides of the square plate 21 .

[0198] Due to the design of the four vertical slots 22, the objects inside the four vertical slots 22 are limited so that they can only move up and down.

[0199] A threaded rod 23 is movably sleeved inside the square plate 21 , and the top and bottom ends of the threaded rod 23 both penetrate the interior of the square plate 21 and extend to the top and bottom ends of the square plate 21 , respectively.

[0200] Since the threaded rod 23 is movably sleeved with the inside of the square plate 21 , the threaded rod 23 can only rotate along the inside of the square plate 21 .

[0201] The outer surface of the threaded rod 23 is threadedly sleeved with a moving block 24 , and the front surface of the moving block 24 is movably connected to the inside of the square plate 21 .

[0202] Since the moving blocks 24 are movably connected to the inside of the square plate 21 , when the threaded rod 23 rotates, the two moving blocks 24 will move in opposite directions.

[0203] Among them, movable blocks 25 are fixedly installed on the left and right sides of the movable block 24, and the movable block 25 is internally movably sleeved with a movable rod 26. There are two movable rods 26, and the other ends of the two movable rods 26 extend to the front of the front of the square plate 21 through the vertical groove 22. The outer surface of the other end of the two movable rods 26 is movably sleeved with an extrusion plate 27.

[0204] When the two moving blocks 24 drive the four moving blocks 25 to move in opposite directions, the four moving blocks 25 will drive the two moving rods 26 to move, and at the same time, the two moving rods 26 will drive the extrusion plate 27 to move backwards.

[0205] A rotating block 28 is fixedly mounted on the top of the threaded rod 23 , and the bottom end of the rotating block 28 is movably connected to the top of the square plate 21 .

[0206] Due to the design of the rotating block 28 , it is convenient for an operator to drive the threaded rod 23 to rotate.

[0207] The working principle and use process of the present invention:

[0208] First, the operator presses the top of the stopper 9. Since the inside of the stopper 9 is movably connected with the fixed block 10, the stopper 9 will rotate with the movably connected part with the fixed block 10 as the axis. At this time, the second spring 11 will be in a compressed state. At the same time, the outer surface of the stopper 9 will cancel the contact with the outer surface of the tooth plate 8 during the rotation. At this time, the stopper 9 will release the blocking effect on the overall movement of the tooth plate 8. Due to the elastic force of the first spring 7, the square block 6 will move under the drive of the first spring 7, and then the square block 6 will drive the telescopic baffle 5 and the tooth plate 8 to move. Due to the design of the interior of the long slot 4, the The movement of the telescopic baffle 5 is limited so that it can only move up and down. At this time, the telescopic baffle 5 will move downward along the inside of the long groove 4 driven by the square block 6. When the bottom end of the telescopic baffle 5 moves below the bottom end of the shield body 1, the overall protection area of ​​the shield body 1 will increase, thereby realizing the function of conveniently extending the shield body 1. At this time, the operator releases the stopper 9. Due to the elastic force of the second spring 11, the stopper 9 will rotate in the opposite direction for resetting. Subsequently, the stopper 9 will contact the outer surface of the tooth plate 8 again during the reverse rotation, thereby fixing the telescopic baffle 5 as a whole.

[0209] Then the operator holds the extrusion plate 27 with one hand, and at the same time, the other hand twists the rotating block 28, at this time, the rotating block 28 will drive the threaded rod 23 to move, because the outer surface of the threaded rod 23 is movably sleeved with the inner part of the square plate 21, at this time, the rotating block 28 will drive the threaded rod 23 to rotate along the inner part of the square plate 21, because the inner parts of the two moving blocks 24 are both threadedly sleeved with the outer surfaces of the threaded rod 23, when the threaded rod 23 rotates, it will drive the two moving blocks 24 to move, because the outer surfaces of the two moving blocks 24 are both movably sleeved with the inner part of the square plate 21, at this time, the two moving blocks 24 will move back to back under the drive of the threaded rod 23, and then the two moving blocks 24 will drive the four moving blocks 25 to move back to back, and then the four moving blocks 25 will drive the two moving rods 26 to move, and at this time, the other ends of the two moving rods 26 will drive the extrusion plate 27 to move backward, and then the extrusion plate 27 will cooperate with the square plate 21 in the backward movement to clamp and fix the operator's hand, thereby realizing the function of preventing the operator from losing the shield during use.

[0210] like Fig. 9 A protection system is installed on the explosion-proof glass and body of the shield, including a multimodal sensor, a controller and an actuator for algorithm operation. A controller for executing the algorithm, as well as sensors and actuators connected to the controller are set on one side of the shield. The sensors include sound sensors, temperature sensors, acceleration sensors, and image sensors; the actuators include vibration motors and voice prompt speakers, and a drive module is installed between the controller and the vibration motor.

[0211] 1. Sensor data input

[0212] Implementing Data: In this step, the system receives real-time data from multiple sensors. Assume that the following sensor data is fed into the system:

[0213] Acceleration sensor: The detected object acceleration is 12.5m / s 2 .

[0214] Image sensor: The object captured is at a distance of 50 meters and a speed of 15m / s.

[0215] Sound sensor: The frequency range of the sound signal detected is between 500Hz and 1500Hz, and the sound intensity is 85dB.

[0216] Infrared sensor: The detected ambient temperature is 32℃, and the local high temperature point reaches 50℃.

[0217] Temperature sensor: The detected ambient temperature change rate is 5℃ / s.

[0218] Result: The above-mentioned sensor data will be integrated into a multi-dimensional data vector for subsequent threat analysis and protection strategy decision-making.

[0219] 2. Data Preprocessing

[0220] Implementation data:

[0221] Sound signal preprocessing: The system performs Fourier transform on the sound signal, converts the time domain signal into the frequency domain signal, and extracts the main frequency components in the spectrum. Assume that after Fourier transform, the extracted frequencies are 700Hz, 1000Hz and 1300Hz, and the corresponding signal energies are 40, 60 and 50 respectively.

[0222] Acceleration signal preprocessing: Smooth the acceleration data and remove noise to keep the acceleration value at 12.5m / s 2 .

[0223] Results: After preprocessing, the frequency distribution of the sound signal, the smoothed result of the acceleration signal, and the filtered signals of the image and temperature will be used as input for subsequent threat analysis.

[0224] 3. Threat Analysis

[0225] Implementation data:

[0226] Self-attention mechanism: Through the self-attention mechanism, the system assigns different weights to different sensor inputs. Assume that the weight of the image sensor input is 0.6, the acceleration sensor is 0.3, and the sound sensor is 0.1.

[0227] Time series analysis: The system analyzes changes in acceleration data and predicts that an object will contact the shield within 3 seconds.

[0228] Result: Threat analysis concluded that the current object has a medium to high threat level, and the object will impact the shield within 3 seconds.

[0229] 4. Threat Level Assessment

[0230] Implementation data:

[0231] Based on the analysis results of different sensors, the system calculates the current threat level L. Assuming that the threat level is represented by a value between 0 and 1, the current evaluation result is L = 0.8.

[0232] Result: The system assesses the current threat level to be 0.8, which is a high threat level. The system needs to respond quickly.

[0233] 5. Reinforcement Learning Strategy

[0234] Implementation data:

[0235] Reward function design: The system gives an immediate reward of 100 in a high-threat situation based on the current threat level L = 0.8.

[0236] The system uses a reinforcement learning algorithm to calculate the best protection strategy and prompts the defender to fully deploy the shield and adjust the angle. Assume that the system's historical data shows that the best protection strategy in this case is maximum protection deployment.

[0237] Results: The optimal protection strategy selected by the reinforcement learning algorithm was: prompting the defender to fully deploy the shield to 100% protection status, adjust the angle to 45°, and protect the approaching object.

[0238] 6. Shield Bearer Tips

[0239] Implementation data:

[0240] Vibration prompt: Assuming the danger level L = 0.8, the vibration proportional coefficient k = 100, the vibration intensity V = 80 is calculated according to the formula. The shield holder will feel a strong vibration prompt.

[0241] Sound prompt: The basic frequency of the sound is 1000Hz, the adjustment coefficient k′=500, and the frequency of the sound prompt is calculated to be 1400Hz according to the formula.

[0242] Result: The system reminds the shield holder through strong vibration prompts and high-frequency alarms that the current threat level is high and the shield holder needs to respond quickly.

[0243] 7. Real-time feedback and adjustment

[0244] Implementation data:

[0245] Assume that the acceleration of the object suddenly slows down at this moment, and the system detects through the sensor that the threat level has dropped to 0.4.

[0246] According to the real-time feedback formula, assuming the learning rate α = 0.5, the system gradually shrinks the shield according to the new threat level.

[0247] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0248] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The attack behavior level identification method based on improved reinforcement learning intelligent algorithm is characterized by: The following steps are involved: The protection system optimizes the protection strategy based on the environmental signals received from different sensors and the application of multimodal sensor fusion, reinforcement learning and real-time feedback control algorithms on the defense equipment. The algorithm adapts to different working conditions and environmental changes, and predicts and warns of future threat status levels.

2. The attack behavior level identification method based on improved reinforcement learning intelligent algorithm according to claim 1 is characterized in that: The specific process is: Step 1: Sensor data collection and preprocessing Infrared sensor: used to detect the thermal radiation signal of the target and detect potential human or explosive danger sources; helps the system to issue an alarm in time when the target approaches, and judges the threat by calculating the temperature difference between the target and the environment; Temperature change formula: ΔT=T target -T ambient Where ΔT is the temperature difference, T target : Target temperature, T ambient : Ambient temperature; When the temperature difference ΔT exceeds the threshold set by the system, the sensor triggers an alarm, indicating the existence of a potential threat; Sound sensor: used to detect sound wave signals in the environment, including gunshots, explosions and abnormal sounds; prompting users to adjust their defensive posture; Fourier transform formula: The Fourier transform converts the time signal x(t) into the frequency domain signal X(f), and the formula is as follows: Among them, x(t): sound waveform, time domain signal; X(f): frequency domain signal, which represents the energy distribution of the sound signal at different frequencies; f: this is the frequency, which represents the different frequency components in the sound signal; e -j2πft :This is the complex exponential function, which is the core of Fourier transform and is used to decompose the time domain signal into different frequency components; here j is the imaginary unit, indicating a complex number; The key frequency components in the sound waves are extracted through Fourier transform and compared with the preset threat template. If the match is high, the system will sound an alarm. Image capture module: used to collect real-time images through the camera, and use the optical flow method to help the system track the trajectory of high-speed moving objects in real time; the system will prompt the user to adjust the defensive posture according to the movement direction and speed of the threatening object; The optical flow method is used to estimate the speed of objects in an image. The formula is as follows: Where, I: brightness value of the image; The rate of change of image brightness in the x and y directions; v x ,v y : The velocity components of the object in the x and y directions; The rate of change of image brightness over time; By solving this equation, the system calculates the object's velocity vector v(x,y), thereby calculating the object's motion path and determining the object's threat level; Acceleration sensor: used to detect changes in acceleration around the defense equipment, especially to identify and detect the impact of high-speed moving objects on the defense equipment, and prompt the defender to adjust the defense status of the defense equipment according to the changes in acceleration; Newton's Second Law: The relationship between the acceleration a of an object and the force F and mass m applied to the object is: Among them, F: force; m: mass of the object; a: acceleration of the object; By measuring the acceleration of an object, the object's trajectory and speed can be inferred; especially when a high-speed object approaches, the acceleration sensor can quickly capture its acceleration changes and issue an early warning; Defense equipment provides appropriate installation locations for sensors. By integrating multiple sensors, the system captures multi-dimensional threat signals in the environment. To ensure the accuracy of the data, the sensor data must first be pre-processed. This includes: Denoising: The sound of wind, background noise signals in sound data, and light noise in images are removed through filters; common methods include Gaussian filtering or median filtering to smooth data and enhance signal reliability; Smoothing filtering: Acceleration and temperature sensor data often experience instantaneous fluctuations; the system smoothes these data through weighted moving average or Gaussian filter to reduce the impact of fluctuations on threat judgment; The weighted moving average formula is: in, is the smoothed data at time t, w i is the weighting coefficient at each time point, x t-1 is the original data at time ti; Feature extraction: Fourier transform is used to extract frequency features from sound signals, while optical flow is used to extract motion features from images. The multimodal data after feature extraction will be input into the subsequent self-attention mechanism for further analysis. Step 2: Data processing using self-attention mechanism The self-attention mechanism is responsible for processing data from infrared, sound, image, and acceleration multimodal sensors. By assigning different weights to different signal sources, the system prioritizes the most important threat signals. This dynamic adjustment mechanism ensures that the system responds to threats in real time and takes protective measures in complex environments. The core of the self-attention mechanism is to calculate the attention weight of each input signal, which is used to judge the importance of the signal and ensure that the system can prioritize the signal with the greatest threat level, which is used to determine the system's protection response. The system receives data x1, x2, ..., x from different sensors n , where each signal x i Represents the input of a sensor; the self-attention mechanism will assign a weight α to each signal i , indicating the priority of the signal; the weight calculation formula is: Where n, j is the number of signals, exp(e i ) means e i The expectation, α i : signal x i The attention weight ranges from 0 to 1, indicating the priority response degree of the signal; i : signal x i The relevance score of the signal x indicates the threat level of the signal to the current situation; the score is determined by the signal x i W is the weight matrix learned by the system during the training process, which is used to measure the impact of each signal source on threat assessment. The Softmax function is used to ensure that the sum of the weights of all signals is 1, which means that the system will normalize all sensor inputs so that different signals can be responded to according to priority; the attention weight is determined by the relevance score e i It is decided that signals with greater threat levels will be given higher weights and thus processed first; Another key feature of the self-attention mechanism is dynamic weight adjustment; as the environment changes in real time, the system adjusts its weights according to changes in different sensor input signals. In particular, when specific threat signals are detected, the system automatically increases the priority of these signals. Signal Input: The system receives input signals from multimodal sensors: x1: The infrared sensor detects a sharp rise in temperature; x2: The sound sensor detects abnormal sound waves; x3: The image sensor recognizes an object approaching at high speed; x4: The acceleration sensor detects the acceleration changes around the defense device; Weight calculation: The system calculates the relevance score e based on the characteristics of each signal i and the attention weight α i ; Dynamic update of weights: According to the formula The system dynamically adjusts the weight of each signal; Dynamic adjustment: When the image sensor detects an object approaching at high speed, the system will automatically increase the priority of the signal, and the defense equipment will be deployed first for protection; the weight setting is updated at a high frequency to ensure that the system can respond to changes in the external environment in real time; When the image sensor detects a high-speed object approaching, the system will respond to this threat first, command the defense equipment to quickly deploy the protection direction to align with the threat source, and expand the baffle to the maximum protection area in a very short time; Infrared sensors continuously monitor changes in ambient temperature. When the temperature rises, the system will adjust its protection strategy to deal with explosion or fire threats. When the sound sensor detects abnormal sound waves, the system will further determine whether the threat comes from gunfire or other abnormal sounds, and make auxiliary judgments; Through the self-attention mechanism, the system dynamically adjusts the priority of sensor data and combines the physical structure of the defense equipment to provide effective protection. Step 3: Time series analysis for sequence data analysis The time series analysis model is the core tool for analyzing time series data. Through the self-attention mechanism, the time series analysis model can identify the patterns in the historical data of the sensor and predict the future threat development trend. This mechanism is particularly suitable for multimodal data processing in dynamic environments. Defense equipment needs to predict threats in advance by analyzing acceleration, image, and sound sensor data. The following is a detailed description of this step. The time series analysis model is constructed by inputting the time series X = [x1, x2, ..., x t ], to predict the output sequence Y = [Y1, Y2, ..., Y t ]; the input sequence X represents the historical data of acceleration, image, sound, and temperature obtained by the system from multiple sensors, and the output sequence Y is the system's prediction of the object's future position, speed, impact time, and future threats; Signal input: The time series analysis model can capture long-term dependencies in the time series, enabling the system to predict future threat signals based on historical data; its processing process includes the following steps: Signal input: The time series X obtained by the system = [x1, x2, ..., x t ], each x i represents the sensor input at a specific time step; x1: data captured by the acceleration sensor 1 second ago; x2: object position data captured by the image sensor 2 seconds ago; x3: comprehensive input of sensors such as image and sound at the current moment; Calculation of query, key, and value matrices: Time series analysis generates query Q, key K, and value V matrices by encoding the input signal at each time step: Query matrix Q: used to predict future state questions; current object acceleration, velocity and other data are encoded into the query matrix to inquire about future threat developments; Key matrix K: saves historical signal features to help identify past data related to the current signal; the historical object motion trajectory is encoded into the key matrix for current matching; Value matrix V: stores the historical data values ​​corresponding to the key matrix for the final prediction calculation; Weighted calculation and output: Time series analysis is done by multiplying the query matrix Q with the transpose of the key matrix K and then scaling it by a factor d k Adjustments are made to calculate the correlation between the current signal and the historical signal; the higher the correlation, the more important the historical signal is to the current prediction; the formula is as follows: Where: Q: represents the question raised by the current input signal when predicting future threats; this matrix is ​​generated by encoding the features of the sensor data in the current time step; K: represents the signal data in the historical time step, which is used to match with the query matrix to help the system determine which historical data is relevant to the current signal; QK T : Calculate the matching degree between the current signal and the historical data; V: represents the eigenvalues ​​of the historical data that matches the key matrix, and the system uses these data to infer the future state; d k : Scaling factor, used to avoid numerical instability caused by excessive input signals; usually the dimension size of the query and key matrices; softmax function: normalizes the calculation results to ensure that the weight distribution is reasonable and the sum of the correlations is 1; This formula calculates the correlation between the query matrix Q and the key matrix K to determine which historical signals best reflect the threat changes in the current environment, thereby using historical data to predict future threat states; by scaling factor d k , the calculation results are stabilized; In the protection system of defense equipment, the time series analysis model analyzes the time series data from multiple sensors to help the system predict future threats in advance; the acceleration sensor and image sensor jointly provide the movement information of the object. The system uses this data to determine the impact time and path of the object and promptly prompt to deploy protection; The accelerometer measures the acceleration of an object and provides historical data about the object's motion state. By analyzing this data, the time series analysis model predicts the object's future position. If the acceleration of an object continues to increase, the system will predict that the object will hit the defense equipment at some point in the future, thereby deploying the defense equipment in advance. The image sensor captures the position changes of the object and combines it with the time series analysis to predict the object's trajectory and final landing point. When the system detects that an object is a high-speed projectile, the time series analysis infers the object's future path through historical data and automatically prompts the defender to adjust the protection direction of the defense equipment to ensure that the protection action is completed before the object contacts the defense equipment. Step 4: Reinforcement learning to optimize defense strategy The core goal of reinforcement learning is to optimize the defense behavior of the protection system by maximizing the cumulative expected reward; the defense strategy of the protection system is π, and its expected reward is defined by the following formula: Among them, J(π): represents the expected return of strategy π, that is, the total cumulative reward obtained after the system executes the strategy; r t : The immediate reward at time step t, which depends on the current defensive behavior and the threat level of the environment; γ: A discount factor ranging from 0 to 1, used to balance the weights of current rewards and future rewards; a larger γ emphasizes long-term rewards, while a smaller γ focuses more on short-term gains; In order to optimize the defense strategy, reinforcement learning uses the policy gradient method to update the policy parameter θ, and its formula is: in, Represents strategy π θ The gradient of θ is used to guide the adjustment of the strategy parameter θ to improve the defense effect of the system; π θ (a t |s t ): The system is in state s t Take action a t The probability of R is determined by the current strategy parameter θ; t : The cumulative reward starting from time step t, reflecting the sum of the defense effects of the system from the current time step to the future time; Through this formula, the protection system continuously adjusts the strategy parameter θ to optimize its defense response, ensuring that the optimal protection strategy is always adopted in different threat scenarios; Specific implementation steps of reinforcement learning to optimize defense strategy Definition of state and action space: In order for the protection system to effectively respond to external threats, it is first necessary to define the state space and action space of the system; State space S: Obtain environmental information through the system's multimodal sensors, including: Acceleration sensor: provides acceleration data of the target object to help determine the movement trend of the object; Image sensor: captures the motion trajectory of the threatening object and determines its position and direction of movement; Sound sensor: detects abnormal sounds in the environment, such as explosions and gunshots, and determines the occurrence of threats; Temperature sensor: monitors ambient temperature and identifies potential fire or explosion threats; Action space A: defines the protection actions taken by the protection system, including: Maximum protection state: indicates that the defense equipment is fully deployed, providing the largest area of ​​protection, suitable for high-threat scenarios; Compact state: prompts the defense equipment to partially retract, maintain basic protection, and save energy, which is suitable for low-threat scenarios; Quickly adjust the angle: Based on the threat signal detected by the sensor, the defender is prompted to adjust the angle of the defense equipment to face the threat object head-on; Reward function design: Reinforcement learning through reward function r t Guide the system to learn the best defense strategy; the design of the reward function should be formulated according to different threat scenarios to ensure that the system can make reasonable protection decisions in complex environments; High Threat Protection Success Reward: When the system successfully defends against high-speed projectiles or explosives, the system will receive a higher immediate reward. t ,encourages the system to adopt the same protection strategy when encountering similar situations in the future; Low-threat energy-saving rewards: In low-threat situations, if the system remains compact and reduces unnecessary defense deployment, positive rewards will be obtained to encourage energy-saving behavior; Defense failure penalty: If the system fails to deploy defense in time, resulting in the threat hitting the defender, the system will receive a negative reward to reduce similar defense failures; Action selection and strategy optimization: At each time step t, the system selects the action based on the current state s. t Choose the best defensive action a t ; The system adopts strategy π θ (a t |s t ) controls the action selection and uses the policy gradient method to update the parameter θ to ensure the maximum cumulative reward in future threat situations; Continuous learning and dynamic adjustment: A key advantage of reinforcement learning is that it can accumulate experience through continuous environmental interaction; over time, the system gradually accumulates more defense experience and continuously optimizes its strategy, enabling it to handle a wider variety of threat scenarios; High Threat Response: When the system detects through sensors that an object is approaching quickly and is highly threatening, the system will prompt the defender to quickly deploy the defense equipment and adjust the angle to maximize the defense against the impact; Low-threat response: In a low-threat environment, the threat detected by the sensor is small, and the system will choose to maintain a compact state, reduce energy consumption and extend the system's operating time; Reinforcement learning combines sensor input and environmental threats to dynamically adjust protection actions in the protection system. Through continuous learning, the system optimizes the deployment status and protection angle of defense equipment based on sensor input information to achieve efficient and intelligent protection. High-threat scenario: When the sensor detects a threatening object approaching at high speed, the system analyzes the threat level through a reinforcement learning algorithm and selects the maximum protection strategy based on historical protection records. At this time, the defender is prompted to quickly deploy the defense equipment to ensure the widest protection area and reduce the risk of attack to the defender. Low-threat scenarios: In low-threat situations, when environmental changes are small and objects move slowly over long distances, the system will maintain a compact state through a low-energy strategy to save energy consumption; this decision guides the system through a reward mechanism, allowing the system to reduce unnecessary protection operations in low-threat situations, thereby extending the working time of the defense equipment.

3. The attack behavior level identification method based on improved reinforcement learning intelligent algorithm according to claim 2 is characterized in that: The defense device is a shield, and the protection system includes a controller set on one side of the shield to execute an algorithm, and sensors and actuators connected to the controller. The sensors include sound sensors, temperature sensors, acceleration sensors, and image sensors; the actuators include vibration motors and voice prompt speakers.

4. The attack behavior level identification method based on improved reinforcement learning intelligent algorithm according to claim 3 is characterized in that: The following steps are also included: Step 5: Real-time feedback and defensive action adjustment The real-time feedback mechanism is regarded as a feedback control system, which is used to make adjustments based on the threat signal input by the sensor and the current defense state of the system. Let the current defense state be S t , the feedback adjustment formula is: S t+1 =S t +α(r t -S t ) Among them, S t : represents the defense state of the system at time t; the system is in different defense states such as fully deployed, defense angle adjustment, and compact state; α: learning rate, ranging from 0 to 1; a larger α value means that the system responds to new threat signals more quickly, while a smaller α value means that the system will gradually adjust and be more adaptable to prevent frequent switching; r t : The feedback signal at the current time t, reflecting the threat level detected by the system from the sensor input information; this feedback signal is based on information such as object acceleration, velocity, and temperature change; S t+1 : represents the defense state of the system at the next time step t+1; When the system detects new threat information, t Will be updated according to the current defense status S t and threat feedback t The formula calculates the difference between the next moment S t+1 Protection adjustment strategy; This feedback formula draws on the feedback control principle in control theory, so that the system can gradually move towards the optimal protection state; Steps to implement the real-time feedback mechanism Sensor data input: The system relies on multi-modal sensors to collect data in real time and assess the threat level based on this data, as follows: Accelerometer: detects objects moving at high speeds, indicating that potential threats are approaching the shield quickly; Image sensor: Identifies the direction, speed, and size of an object, helping the system determine the threat level of the object; Sound sensor: detects abnormal sounds and indicates that the shield protection system is facing threats from sound waves; Temperature sensor: monitors drastic changes in ambient temperature, high temperature explosion or fire source, and prompts the system to face extreme danger; Real-time adjustment of defense status: The system adjusts the defense status based on the threat level signal fed back by the sensor and the current defense status S t , using formulas to prompt the defender to dynamically adjust the shield status; specific adjustments include: Fully deploy the shield: When the threat level is high, the system will prompt the operator to fully deploy the shield to provide maximum protection and prevent high-speed approaching objects from hitting the shield holder; Compact State: In low-threat situations, the system keeps the shield in a compact state, saving energy and avoiding unnecessary deployment operations; Protection angle adjustment: When the sensor identifies a threatening object coming from a specific direction, the system will prompt the operator to adjust the shield angle to face the threatening direction at the best angle to reduce the exposed area; Real-time response and learning: Through real-time feedback mechanism, the system can continuously adjust the shield status to cope with the dynamically changing threat environment; reinforcement learning strategy supports the system to gradually optimize the protection strategy in different scenarios; Quickly adjust the protection angle: When the sensor detects an object approaching at high speed from a certain angle, the system will quickly prompt the protector to adjust the protection angle and deploy the shield accordingly; Gradually optimized protection strategy: In the ever-changing threat scenarios, the system gradually optimizes the protection action through sensor feedback signals to ensure that the shield holder is always in the best protection state; Step 6: Shield Bearer's Prompt System The shield holder's reminder system is a core module of the shield protection system. It reminds the shield holder of the current defense status and danger level through vibration and sound prompt feedback. This design ensures that the shield holder can quickly and accurately take corresponding defense measures when facing complex and dynamic threats. The following is a detailed implementation plan, and gives specific implementation cases and descriptions of related parameter variables; The prompt system provides feedback to the shield holder through the intensity of vibration prompts and sound prompts based on the detected danger level L; as the danger level changes, the feedback intensity will also be adjusted to ensure that the shield holder can take appropriate actions according to different danger situations; the vibration prompt intensity formula is: V=kL Where V: the intensity of the vibration prompt, which indicates the vibration feedback received by the shield holder; as the danger level increases, the intensity of the vibration prompt also increases; L: the danger level detected by the current system, ranging from 0 to 1, 0 represents no threat, and 1 represents the highest threat; k: the proportional coefficient, used to adjust the sensitivity of the vibration prompt; a higher k value means that the vibration intensity will increase significantly when the danger level changes slightly; the sound prompt frequency formula is: f=f0+k ′ L f: The frequency of the sound prompt, which is used to remind the shield holder of the current threat level through the sound alarm; as the danger level increases, the frequency of the sound prompt will also increase; f0: The basic frequency, which indicates the frequency of the sound prompt when there is no threat; the shield holder perceives the change of danger level through the change of sound frequency; k ′ : Adjustment coefficient, which determines the influence of danger level on the frequency of sound prompts; higher k ′ The value means that when the danger level increases, the frequency change of the sound prompt becomes more obvious; Implementation steps of the shield bearer reminder system Sensor data input: The prompt system obtains the current danger level L from multiple sensors and feeds it back to the shield holder in real time; the following are the relevant sensor inputs: Acceleration sensor: detects the speed and acceleration of an object to determine whether a threatening object is approaching quickly; Image sensor: captures the movement trajectory, direction and distance of threatening objects to help the system determine the threat level of the object; Sound sensor: detects abnormal sounds in the environment and indicates potential threats; Temperature sensors: monitor ambient temperature changes, especially when detecting fire or explosives; Vibration prompt: When the system detects a threat, it provides feedback to the shield holder through vibration prompt; the vibration intensity is adjusted according to the danger level L: When the threat level is high: the system will send a strong vibration prompt to the shield holder, reminding the shield holder to take defensive measures immediately; At low threat level: the system will only provide slight vibration to remind the shield holder that the current threat is low; at this time, the V value is small and the vibration feedback is limited; Sound prompt: At the same time, the system issues an alarm through the sound prompt module, and the frequency f of the sound prompt increases as the danger level L increases: When the threat is high: the frequency of the sound prompt increases significantly, reminding the shield holder that he is facing a serious threat; as the danger level L increases, the sound becomes more urgent; When the threat is low: the sound prompt frequency is kept at a low level f0, reminding the shield holder that he is in a relatively safe state; Prompt feedback and defensive action adjustment: The prompt system is combined with the shield's defensive actions to ensure that the shield holder can quickly make corresponding adjustments based on the feedback provided by the prompt system; when the shield holder receives high-threat vibration and sound prompts, defensive measures are taken immediately.

5. A foldable shield that is easy to extend, characterized in that: Included are: A shield body (1), wherein explosion-proof glass (2) is fixedly installed inside the shield body (1), a square groove (3) is provided at the bottom end of the front side of the shield body (1), and a long groove (4) is provided at the bottom end of the shield body (1); An extension mechanism, the extension mechanism being arranged inside the long slot (4); The extension mechanism comprises a telescopic baffle (5), the outer surface of the telescopic baffle (5) is movably connected to the inside of the long slot (4), a square block (6) is fixedly installed on the top of the telescopic baffle (5), the outer surface of the square block (6) is movably connected to the inside of the long slot (4), a first spring (7) is fixedly installed on the top of the square block (6), the top of the first spring (7) is fixedly connected to the inside of the long slot (4), a tooth plate (8) is fixedly installed on the front of the telescopic baffle (5), and the front of the tooth plate (8) A stopper (9) is movably connected, the other end of the stopper (9) extends to the front of the shield body (1) through the square groove (3), the outer surface of the stopper (9) is movably connected to the inside of the square groove (3), a fixing block (10) is movably sleeved inside the stopper (9), the back side of the fixing block (10) is fixedly connected to the front side of the shield body (1), a second spring (11) is fixedly mounted on the top of the back side of the stopper (9), and a rear end of the second spring (11) is fixedly connected to the front side of the shield body (1).

6. The foldable shield that is easy to extend according to claim 5, characterized in that: Round rods (12) are fixedly mounted on both left and right sides of the shield body (1); round tubes (13) are movably sleeved on the top and bottom ends of the outer surfaces of the round rods (12); a folding baffle (14) is fixedly mounted on the outer surface of the round tube (13); and the back surface of the folding baffle (14) is movably connected to the front surface of the shield body (1).

7. The foldable shield that is easy to extend according to claim 1, characterized in that: A fixing plate (15) is fixedly mounted on the top of the front face of the shield body (1), vertical rods (16) are fixedly mounted on both left and right sides of the top of the fixing plate (15), and a long rod (17) is movably sleeved on the outer surface of the vertical rod (16); A third spring (18) is fixedly mounted on the bottom end of the long rod (17), and the bottom end of the third spring (18) is fixedly connected to the top end of the fixed plate (15); A first round block (19) is fixedly mounted on both left and right sides of the long rod (17); the bottom end of the first round block (19) is movably connected to a second round block (20); the bottom end of the second round block (20) is fixedly connected to the top end of the circular tube (13); and the bottom end of the second round block (20) is movably connected to the top end of the circular rod (12).

8. The foldable shield that is easy to extend according to claim 5, characterized in that: A square plate (21) is fixedly mounted on the front of the shield body (1), and vertical grooves (22) are provided at both upper and lower ends of the left and right sides of the square plate (21); A threaded rod (23) is movably sleeved inside the square plate (21), and the top and bottom ends of the threaded rod (23) both penetrate the interior of the square plate (21) and extend to the top and bottom ends of the square plate (21) respectively; The outer surface of the threaded rod (23) is threadedly sleeved with a moving block (24), and the front surface of the moving block (24) is movably connected to the inside of the square plate (21); The left and right sides of the moving block (24) are both fixedly mounted with movable blocks (25), the interior of the movable block (25) is movably sleeved with a moving rod (26), the number of the moving rods (26) is two, the other ends of the two moving rods (26) extend to the front of the square plate (21) through the vertical slot (22), and the outer surfaces of the other ends of the two moving rods (26) are movably sleeved with an extrusion plate (27); A rotating block (28) is fixedly mounted on the top end of the threaded rod (23), and the bottom end of the rotating block (28) is movably connected to the top end of the square plate (21).

Images